If you’ve been keeping up with AI advancements lately, there’s a huge chance you’ve heard about him. Yann LeCun, known as one of the "Godfathers of AI," has made major developments in Meta as its chief AI scientist. He among other known figures have been playing a huge part in revolutionizing the technology we know today.
But what are the things you might not know about Yann? From his development of convolutional neural networks (CNN) to his newly proposed Image Joint Embedding Predictive Architecture (I-JEPA), we will go through everything you need to know about Yann LeCun's major contribution to AI advancements as well as his thoughts on what AI could become in the future.
Who Is Yann LeCun?
Yann André LeCun is a computer scientist from France who pioneered machine learning, computer vision, mobile robotics, and computational neuroscience. Yann is widely recognized for his contributions to convolutional neural networks (CNN) and is regarded as one of the convolutional net's pioneers.
He received his Electrical Engineer Diploma from ESIEE in Paris in 1983 and his PhD in Computer Science from Universite Pierre et Marie Curie (also in Paris) in 1987. Notably, during his PhD, he proposed an early version of the back-propagation learning technique for neural networks, which I’ll explain in detail later.
He began teaching at New York University in 2003 and founded the NYU Center for Data Science a few years later in 2012.
Shortly after in 2013, LeCun was hired by Facebook to manage its newly formed AI research branch. He is currently the Chief AI Scientist at Facebook and is also involved in Meta. He’s been at the company for 10 years.
His Massive Contributions To Meta
Yann is a major figure at Meta, since his appointment in 2013, the company has made significant advances in artificial intelligence, particularly in natural language processing and computer vision. “Despite significant advances in AI research, we are still a long way from creating machines that can think and learn like humans”, he said.
Yann draws a clear distinction between a teenager learning to drive in about 20 hours with no prior experience behind the wheel and current autonomous driving systems that need huge amounts of labeled training data and reinforcement learning trials to operate yet still fall short of human driving reliability.
He recently proposed a new architecture Intended to overcome key limitations of even the most advanced AI systems today. The Image Joint Embedding Predictive Architecture (I-JEPA). It is a model that learns by creating an internal understanding of the world, rather than comparing pixel-level details. I-JEPA not only succeeds in numerous computer vision tasks, but it also outperforms commonly used models in terms of computational efficiency.
At a high level, the I-JEPA goal is to predict the representation of some parts of an input (such as an image or text) based on the representation of other parts of the same input. It is intended to avoid the biases and difficulties associated with another widely used method known as invariance-based pretraining because it does not involve collapsing representations from many views/augmentations of an image to a single point.
What's amazing is that the representations obtained using I-JEPA are adaptable to a broad range of applications without requiring substantial fine-tuning.
His Thoughts on Job Employment
While many are concerned about the potential of AI to replace a wide range of jobs as well as some companies, he assured that It’s not likely to happen. He told the BBC in an interview, "This is not going to put a lot of people out of work permanently". “But work would change because we have what the most prominent jobs will be 20 years from now”, he said. “Intelligent computers would create ‘a new renaissance for humanity’ the way the internet or the printing press did”.
He also spoke ahead of a vote on Europe's AI Act which is designed to regulate artificial intelligence. According to his conversations with European AI start-ups, "they don't like it at all, they think it's too broad, maybe too restrictive." However, he stated that he was not an expert on the legislation. He said that he is not opposed to regulation, but that each application would require its own set of laws, for example, separate rules would govern AI systems in cars and those scanning medical photos.
His Thoughts on LLMs
Yann provides his thoughts regarding Language Language Models (LLMs), pointing out the buzz surrounding them. He urges to be careful, pointing out that simply increasing the size of these systems would not drive us to human-like intelligence. He describes the progress as "interesting," but not as "the final destination."
He also stresses that open-source generative models have existed for a long time without producing the catastrophic issues that many are afraid of. He believes that possibilities of mass disinformation and hacking are excessively dramatic, similar to something out of a James Bond film.
Yann believes that society should trust technology to be used for good, even if there is a risk of misuse. He draws analogies between past technologies such as the printing press and the internet, emphasizing their huge benefits despite certain drawbacks. He follows by stating that, while some may sense existential threats, current AI systems do not.
His Contribution to Neural Networks
His work on convolutional neural networks (CNNs) has had a significant impact on computer vision, revolutionizing tasks such as picture recognition, object detection, and segmentation. Particularly, his achievements garnered him the prestigious “Turing Award” in 2018, which has been compared to the "Nobel Prize of Computing," which he shared with Yoshua Bengio and Geoffrey Hinton, highlighting the overall effect of their work on deep learning.
It’s also great to mention that he co-created the DjVu image compression technique with Léon Bottou and Patrick Haffner, which is specifically geared towards the compression of scanned documents in color at high resolution. It allows any screen with an Internet connection to access and display images of scanned pages while correctly recreating the fonts, color, drawings, images, and paper texture. A typical magazine page in color at 300dpi can be compressed down to 40 to 60 KB, which is approximately 5 to 10 times better than JPEG for a similar level of subjective quality.
He also co-created the Lush programming language (with bottou) which is a Lisp-like object-oriented programming language designed for researchers, experimenters, and engineers interested in numerical applications, including computer vision and machine learning.
LeCun is widely regarded as a pioneer in the fields of computing and artificial intelligence. He has been honored with several prestigious awards in recognition of his significant contributions to various fields such as:
- Turing Award (2018): Often hailed as the "Nobel Prize of Computing," this annual recognition by the Association for Computing Machinery (ACM) lauds individuals for their profound contributions to the computing realm.
- AAAI Fellow (2019): Acknowledged by the Association for the Advancement of Artificial Intelligence (AAAI), this fellowship underscores LeCun's significant contributions to the field of artificial intelligence.
- Legion of Honour (2020): Bestowed by the French government, this esteemed order of merit recognizes LeCun's outstanding military and civil contributions.
- Doctorates Honoris Causa: Multiple universities have awarded LeCun honorary degrees in recognition of his remarkable contributions to specific fields and society at large.
- Pender Award: Presented by the University of Pennsylvania, this accolade celebrates LeCun's substantial contributions to electrical engineering.
- Holst Medal: Conferred by the Technical University of Eindhoven and Philips Labs, this medal acknowledges LeCun's significant impact on the field of electrical engineering.
- Nokia-Bell Labs Shannon Luminary Award: Recognizing outstanding contributions to information theory, this award from Nokia Bell Labs highlights LeCun's influential work.
- IEEE PAMI Distinguished Researcher Award: Bestowed by the Institute of Electrical and Electronics Engineers (IEEE), this award celebrates LeCun's noteworthy contributions to pattern analysis and machine intelligence.
- IEEE Neural Network Pioneer Award: Presented by the IEEE Computational Intelligence Society, this award honors LeCun's substantial impact on the field of neural networks.
What He Thinks About AI
Yann's view is mixed with both personal and professional experience. Despite his previous successes with neural networks, he does not allow them to take over his confidence in current controversial opinions. As a scientist, he’s also willing to express what he thinks.
His time at Meta has given him a unique look behind the scenes, allowing him to see motivations that may not match public perceptions. When confronted with claims that working at Meta is inherently unethical, he responds by highlighting the complicated nature of ethical considerations in technology.
He also said that the rise of artificial general intelligence (AGI) within the next five years, is unlikely. He does, however, believe that machines will eventually outperform human intelligence. He also emphasizes that this development should not be interpreted as a threat.
Even as someone who is referred to as one of the “Godfathers of AI” he understands the risk of unforeseen consequences in the field of artificial intelligence. Still, he’s pretty determined to proactively take on and navigate ethical challenges that come with studying artificial intelligence, always striving to chart the best path forward. He’s an influential figure in the industry for sure, and it’ll be exciting to see what he does in years to come.